Smart Pasture Welfare Monitoring System Based On Multi-Source Heterogeneous Data Fusion

Combining the digital and intelligent methods of modern information technology with pasture management can help to realize the integration of cattle welfare control in pastures.At present,most studies on cattle welfare monitoring systems focus on behavioral aspects,often neglecting the influence of basic factors such as farm environment and feeding,which leads to a homogenization of correlation information and affects the reliability of welfare assessment.To address these problems,this paper investigates and designs a pasture welfare monitoring system based on multi-source heterogeneous data,based on monitorable information that can directly reflect the cattle welfare status,such as ear and coat color,as well as environmental and feeding data.The main work is as follows:(1)Construction of a cattle ear tracking discrimination model.Based on Deep Lab Cut model,we track and match the key points of cattle ears to get the coordinates of key points of cattle ears;on this basis,we design the cattle ear relative fluctuation algorithm and fluctuation judgment algorithm to get the relative distance and fluctuation characteristic quantity of cattle ears,and classify the fluctuation situation of cattle ears according to their standard deviation size.The experimental results show that the best effect is achieved by selecting Res Net-50 as the backbone network,and the detection speed reaches 20.12f/s with the smallest misjudgment error for cattle ear behavior tracking.(2)Construction of cattle hair color classification model.The base model of YOLOv5 is selected,and the cattle hair color information is obtained by different scale adaptive fusion methods,and different attention mechanisms and adaptively spatial feature fusion(ASFF)are embedded to obtain the improved model of YOLOv5.The experimental results show that the model finally achieves an average accuracy of up to 95.36% and a monitoring frame rate of 52.23 f/s when SENet and CBAM and ASFF are embedded in YOLOv5,and its average accuracy is 10.01 percentage points higher than that of the original model.Improving the accuracy of recognizing cattle hair color features.The effectiveness of the improved model is verified through experiments.(3)Construct a pasture welfare evaluation model.A fuzzy neural network model was designed to input cattle ear fluctuation information,hair color classification information,and environmental and feeding data into three parallel groups of BP neural networks to obtain various categories of grades,which were then converged into a fuzzy controller for fuzzy logic inference to output pasture welfare grades.The accuracy of the welfare evaluation model is verified through experiments.(4)Develop a smart ranch welfare monitoring system and embed the above model into the system to realize the monitoring function of the ranch welfare system.The effectiveness of the system function is tested through experiments.